Speakers
Description
Background: Adverse pregnancy outcomes (APOs) remain a major public health issue, especially in low- and middle-income countries (LMICs). Machine learning (ML)-based risk prediction models present opportunities for early identification and intervention, yet there is limited evidence of their application and predictive performance in LMICs. The review aimed to map the existing evidence on ML models and input features used to predict APOs in LMICs.
Methods: This review was guided by the Joanna Briggs Institute methodology for scoping reviews and adhered to the Preferred Reporting Items for Systematic Reviews and Meta-analysis extension for Scoping Reviews. A comprehensive literature search was conducted in PubMed, Cochrane Library, Web of Science and Scopus for articles from January 1, 2000, to June 26, 2024.
Results: Our search strategy yielded 4,680 records from which 351 duplicates were removed. After titles and abstracts screening, 114 full-text articles were assessed for full-text screening, out of which 25 were selected for inclusion in the review. An additional nine articles were identified from the references of the included studies, resulting in 34 being included in the final review. All the ML models used across the studies were supervised learning. The features most commonly used to train the ML models comprised maternal characteristics, clinical and obstetric history.
Conclusion: This review highlights the evolving yet limited application of ML-based risk prediction models for APOs in LMICs. Validating these models across different populations may be crucial for their integration into routine clinical care, ultimately enhancing maternal and child health.